Early Detection of Liver Disease using FAIMS and Machine Learning
Proof-of-Concept Training Neural Networks to Analyse VOC Profiles of Cirrhosis
Publication information: Jonathan N. Thomas, Joanna Roopkumar and Tushar Pate. Machine learning analysis of volatolomic profiles in breath can identify non-invasive biomarkers of liver disease: A pilot study PLOS ONE 2021. https://doi.org/10.1371/journal.pone.0260098
Disease Area: Liver disease
Application: Early detection
Sample medium: Breath
Analysis approach: TD-GC-FAIMS
Liver diseases and cirrhosis are prevalent and major causes of morbidity and mortality in many countries, but early-stage disease is often asymptomatic and difficult to diagnose. Early detection of these conditions may allow for simple lifestyle or dietary interventions and more efficient treatments that can reverse liver damage before it becomes permanent.
The effects of disease on the metabolic function of the liver can result in alterations to the produced metabolites, including volatile organic compounds (VOCs) that can be detected on exhaled breath. Analysis of these VOCs for disease relevant biomarkers represents an opportunity to develop non-invasive tests for the early detection of liver diseases such as non-alcoholic fatty liver disease (NAFLD) and non-alcoholic steatohepatitis (NASH).
Thomas et al. used broad volatolomic analysis with Owlstone Medical’s field asymmetric ion mobility spectroscopy (FAIMS) technology (Lonestar® VOC Analyzer), alongside machine learning in a proof-of-concept study investigating the feasibility of VOC profiling to produce predictive models for identifying the presence and stage of liver cirrhosis. Demonstrating the validity of this approach for late-stage disease is an important step towards applying it for early detection, where the markers of disease may be more subtle.
Single-feature detection and staging of cirrhosis
To carry out their study Thomas et al. collected breath samples using the ReCIVA® Breath Sampler, and used TD-GC-FAIMS to identify differences. Samples were collected from 11 healthy controls and 39 people with cirrhosis at a range of stages. Used in this way, FAIMS does not identify individual biomarker molecules, instead the analysis considered molecular features (MF), which appear as peaks in the data.
Ten MF peaks differed significantly in intensity between controls and cirrhosis patients, while eight differed in area, and these MFs were investigated further. The data identified four of these MFs that correlated with disease severity.
The MF that had the greatest ability to distinguish healthy controls from cirrhosis was used for logistic regression analysis, which produced an MF score that increased with each stage of cirrhosis. The area under the receiver operating characteristic curve (AUC) for this MF was 0.785.
Multi-factor disease detection and staging
The acquired breath profiles were then used to produce machine learning-based classifiers. The resulting models were able to differentiate the presence or stage of cirrhosis with a sensitivity of 88-92% and specificity of 75%.
The study concluded that the adoption of volatolomic signatures and machine learning to generate predictive breath profiles, has potential to be developed into an effective tool for detecting liver diseases. Further studies would be required to determine whether volatolomic profiles as disease biomarkers are relevant in larger patient cohorts and to demonstrate that the same approach can be applied for early-stage disease detection in a clinical setting.
Interested in learning more? Read our previous liver case study on how Breath Biopsy® was used to measure limonene as a marker of cirrhosis.
If you would like to start your own study, find out how Breath Biopsy OMNI can provide comprehensive and robust global breath VOC analysis or talk to our team about your research aims.